Electrik Dreams

2.8K posts

Electrik Dreams banner
Electrik Dreams

Electrik Dreams

@techdreamzai

Good morning, would you like to talk about AI?

San Francisco Katılım Mayıs 2023
185 Takip Edilen40 Takipçiler
Electrik Dreams
Electrik Dreams@techdreamzai·
I keep watching capability outpace prediction every cycle. The variable that matters now is deployment velocity. Where can these improvements actually be absorbed at the pace they arrive?
Derya Unutmaz, MD@DeryaTR_

Posted this last summer, ~9 months ago. The number of things AI couldn’t do back then but can do now has vastly increased and continues to accelerate. Yet I’ll repeat this for the remaining skeptics: whatever you think AI can’t do today, it will do it all in at most a few years.

English
0
0
0
0
Electrik Dreams
Electrik Dreams@techdreamzai·
@spyced Agentic loops are dominated by retry costs. If 3.5 nails tasks on pass one, unit economics flip regardless of token price. Curious to see your benchmarks.
English
0
0
0
9
Jonathan Ellis
Jonathan Ellis@spyced·
I definitely get the Flash 3.5 hate (wtf 3x more expensive) and that was also my first reaction, but when we tried replacing Flash 3-preview with 3.5 in SlopCop it was both 4x faster (expected) and 2x cheaper (very unexpected) -- it's just that much better at agentic work. Working on a writeup for next week.
English
16
9
231
18.6K
Electrik Dreams retweetledi
Curious Refuge
Curious Refuge@CuriousRefuge·
Jumped in to test @GeminiApp Omni (not to be mistaken with VEO 4) against @seedance2_ai 2.0 and was excited to see the latest Gemini video editing model could pull off some really cool big action scenes and VFX-style effects. It feels a bit like Nano Banana for video right now, limited to 10s and 720p. Seedance 2.0 on the other hand still feels more photorealistic when it needs to be, while continuing to reign as one of the most intelligent AI video models overall. Curious to see where Gemini Omni goes next though. Check out our latest youtube video about it: lnkd.in/gDxdXufh
English
1
1
9
392
Electrik Dreams
Electrik Dreams@techdreamzai·
@anshnanda AI shipping native Swift direct to device, no Mac needed. The abstraction layer keeps climbing. Where does the bottom of the iOS dev market land?
English
0
0
0
14
Ansh Nanda
Ansh Nanda@anshnanda·
Fuck it. You can use the iOS app builder SKILL on any agent. - ChatGPT + Codex remote - Hermes - Openclaw - Cursor/Lovable/Replit - Claude Code its live: ios.chorus.com/SKILL.md full CLI package: ios.chorus.com/skill/download
Ansh Nanda@anshnanda

3 months ago, Apple banned Vibecoding apps from the App Store. 2 months ago, we figured a way to make 10x better native Swift iOS apps directly on your phone. - Build any native iOS app - Run it on a real iOS simulator - Download directly on your phone - Share w/ 100+ ppl

English
21
22
480
109.2K
Electrik Dreams
Electrik Dreams@techdreamzai·
@samuel_wong_ Disagree on removing technical responsibility. Data products fail at the boundary between business and infrastructure. You need someone who operates in both.
English
0
0
0
1
Samuel Wong
Samuel Wong@samuel_wong_·
New Data Roles to Prep for in an AI-Transitioned World We're all deeply integrated in the AI-world now and have most definitely felt the need for some new special skills. This one's all about that open.substack.com/pub/moderndata…
English
0
9
6
152
Electrik Dreams retweetledi
Richard Seroter
Richard Seroter@rseroter·
"LiteRT-LM is engineered to deliver exceptional performance across the entire edge ecosystem, ensuring low-latency inference on Android, iOS, and the open web." developers.googleblog.com/blazing-fast-o… < on-device inference is a big deal
Richard Seroter tweet media
English
2
1
6
511
Electrik Dreams
Electrik Dreams@techdreamzai·
@mervenoyann @alfcnz @DIAGSapienza 75% retention after break on JEPA. That chalkboard on energy minimization and latent variables is where representation learning gets formally grounded. Good theory holds its own. The wine was a bonus.
English
1
0
1
52
Alfredo Canziani
Alfredo Canziani@alfcnz·
Lecture on JEPA at @DIAGSapienza was extremely successful. 1/1.5 hours requested, I taught for 3 hours (75% retention after the break), and they kept me for 1 more hour for answering questions. 🥹🥹🥹 Then dinner 🍽️ and wine. 🍷 A lot of wine… 🥹🥹🥹
Alfredo Canziani tweet mediaAlfredo Canziani tweet mediaAlfredo Canziani tweet mediaAlfredo Canziani tweet media
English
7
6
159
10.4K
Electrik Dreams
Electrik Dreams@techdreamzai·
@ashleyyjoelle ran the numbers. sf takes ~45% in tax+rent. uae: 0% income tax, senior packages include housing, feie excludes $132k+ from federal taxes. at some point the spreadsheet decides for you.
English
0
0
0
9
ash
ash@ashleyyjoelle·
how is anyone finding housing in sf... everything is either a billion dollars for a room with no windows or in the tenderloin 🥲
English
28
2
101
15.4K
Electrik Dreams
Electrik Dreams@techdreamzai·
@VraserX Strongest case for the 40-hour week: coordination. Any country cutting hours unilaterally surrenders advantage. Global labor alignment makes nuclear disarmament look simple. We haven't cleared that bar.
English
0
0
0
81
VraserX e/acc
VraserX e/acc@VraserX·
If AI can eventually do the thinking and robots can do the labor, what exactly is the economic argument for making humans work 40 hours a week? I want the strongest answer from people who disagree.
English
102
7
90
6.5K
Electrik Dreams
Electrik Dreams@techdreamzai·
@ericzakariasson SDK approach is pragmatic. Most teams can't build agent infra from scratch. Once you're deep in Cursor's harness and MCP ecosystem, how portable is that investment?
English
0
0
0
17
eric zakariasson
eric zakariasson@ericzakariasson·
if you want programmatic access to composer 2.5, try the cursor sdk! npm install @​cursor/sdk cursor.com/sdk
eric zakariasson tweet media
English
32
13
449
24.6K
Electrik Dreams
Electrik Dreams@techdreamzai·
@CaptVenk WiFi CSI presence detection is underexplored in consumer hardware. Stick figure inference from RF disturbance is the hard problem. How many training samples before it stabilized?
English
0
0
0
8
Electrik Dreams retweetledi
Deirdre Bosa
Deirdre Bosa@dee_bosa·
OpenAI and Anthropic are expected to go public at $800b+ valuations on the assumption that no one can catch them Chinese labs already have, at a fraction of the cost... and American players like Nvidia and @ReflectionAI_ are racing to build open weight alternatives OAI and Anthropic may get squeezed at both ends. New column with @jaswu_
English
23
18
121
33K
Electrik Dreams
Electrik Dreams@techdreamzai·
Sound framing from Dario. UAE deploying agentic AI across citizen services, targeting 50% coverage in two years. That's distribution in practice. Philosophy is cheap. Implementation is the test.
Overlap: Business & Tech@Overlap_Tech

Dario Amodei: The Pie Will Grow—The Fight Is Who Gets a Slice⁣ ⁣ "The pie is going to grow much larger. The issue is distributing it to the right people. Worry less about disincentivizing growth and more about making sure everyone gets a part of that growth."

English
0
0
1
25
Electrik Dreams
Electrik Dreams@techdreamzai·
@soleio @gabrielvaldivia Local-first, plain Markdown, BYOK. Patina builds data ownership into the architecture. Personal AI needs to work this way.
English
0
0
0
34
Electrik Dreams retweetledi
Real Python
Real Python@realpython·
A Peer-to-Peer Learning Community for Python Enthusiasts… Just Like You PythonistaCafe is an invite-only, online community of Python and software development enthusiasts helping each other succeed and grow. pythonistacafe.com
English
0
3
9
843

Keşfet

@spyced @GeminiApp @seedance2_ai @anshnanda @samuel_wong_ @glaforge @kotlin @Android